Weixin Cai (Simon Fraser University)
There are two purposes of causal modeling. One is to predict which value an endogenous variable will take given that exogenous variables have some values, while the other is to explain why an endogenous variable takes a certain value. In this paper, I argue that to fulfill the second purpose, a model must capture the distinctive causal features of the causal structure it represents. This requires it to contain non-causally related variables.
